[KDD'2026] "VideoRAG: Chat with Your Videos"
Python large-language-modelsllmslong-video-understandingmulti-modal-llmsragretrieval-augmented-generation
make llm module configurable in videorag_expiriment
Gerald Hewes9e2c229910f7
diff --git a/longervideos/videorag_experiment.py b/longervideos/videorag_experiment.pyindex 1c7067a..99e15a8 100644--- a/longervideos/videorag_experiment.py+++ b/longervideos/videorag_experiment.py@@ -15,6 +15,7 @@ import argparseparser = argparse.ArgumentParser(description="Set sub-category and CUDA device.")parser.add_argument('--collection', type=str, default='4-rag-lecture')parser.add_argument('--cuda', type=str, default='0')+parser.add_argument('--config', type=str, choices=['openai', 'azure', 'ollama'], default='openai')args = parser.parse_args()sub_category = args.collection@@ -24,6 +25,14 @@ os.environ["OPENAI_API_KEY"] = ""from videorag._llm import openai_config, azure_openai_config, ollama_configfrom videorag.videorag import VideoRAG, QueryParam+config_map = {+ 'openai': openai_config,+ 'azure': azure_openai_config,+ 'ollama': ollama_config+}++llm_config = config_map[args.config]+if __name__ == '__main__':multiprocessing.set_start_method('spawn')@@ -31,14 +40,14 @@ if __name__ == '__main__':video_base_path = f'longervideos/{sub_category}/videos/'video_files = sorted(os.listdir(video_base_path))video_paths = [os.path.join(video_base_path, f) for f in video_files]- videorag = VideoRAG(llm=ollama_config, working_dir=f"./videorag-workdir/{sub_category}")+ videorag = VideoRAG(llm=llm_config, working_dir=f"./videorag-workdir/{sub_category}")videorag.insert_video(video_path_list=video_paths)## inferencewith open(f'longervideos/dataset.json', 'r') as f:longervideos = json.load(f)- videorag = VideoRAG(llm=ollama_config, working_dir=f"./videorag-workdir/{sub_category}")+ videorag = VideoRAG(llm=llm_config, working_dir=f"./videorag-workdir/{sub_category}")videorag.load_caption_model(debug=False)answer_folder = f'./videorag-answers/{sub_category}'